supplemental material for boundary in warm dense hydrogen ... · simulation shifts the imt boundary...
TRANSCRIPT
-
1
Supplemental Material for
“Fully Consistent Density Functional Theory Determination of the Insulator-Metal Transition
Boundary in Warm Dense Hydrogen”
Joshua Hinz1, Valentin V. Karasiev1*, S. X. Hu1, Mohamed Zaghoo1, Daniel Mejía-Rodríguez2,
S.B. Trickey2, and L. Calderín3
1 Laboratory for Laser Energetics, University of Rochester, Rochester, NY 14623
2 Quantum Theory Project, Department of Physics, University of Florida, Gainesville, FL 32611
3 Department of Materials Science and Engineering, University of Arizona, Tucson AZ 85721
06 Feb. 2020
Ab initio simulations:
Born-Oppenheimer molecular dynamic (BOMD) simulations were performed via the Vienna ab
initio simulation package (VASP) [1-3] using the SCAN-L [4,5] exchange-correlation (XC)
functional. The rVV10 [6,7] correlation correction was included in all simulations, unless
specifically stated otherwise. The motive is to incorporate the long range van der Waals
interaction that are not represented well by SCAN (or SCAN-L), despite its reasonable treatment
in the vicinity of equilibrium bond lengths. Those long-range contributions may be important
role in H2 dissociation. All simulations used the NVT ensemble on a system of 256 atoms in a
cubic super cell, with lattice parameters ranging from 7.05 to 10.27 Å. Electronic structures
were calculated at the Γ point with 256 bands. Convergence tests, see Figure S1, indicate that use
of Γ-point-only sampling introduces a maximum pressure uncertainty of 3%. Each MD
trajectory consists of 5000 to 6000 steps with a time step of 0.1 fs, giving a total simulation time
of 0.5 to 0.6 ps. The bare ion Coulomb potential was treated via the projector augmented wave
(PAW) framework [8] with the use of PBE PAWs (metaGGA PAWs are not available in VASP
at present).
Ring polymer path integral MD (PIMD) calculations for inclusion of nuclear quantum
effects (NQEs) were done with the i-PI [9,10] interface to Quantum Espresso (QE) [11,12].
Those simulations used the SCAN [13] XC functional with the rVV10 van der Waals correction.
These calculations used a local pseudo-potential [14] while the rest of the technical parameter
mailto:*[email protected]
-
2
values were as consistent possible with those used in the BOMD simulations. All MD simulation
parameters are tabulated in Table S1 for reference.
With the exception of the 2500 and 3000 K IMT points, all other transition points were
obtained from simulations along various isochores with densities ranging from 0.8 to 1.15 g/cm3.
This procedure was chosen so as to follow thermodynamic paths consistent with those in the
static compression experiments. As a technical aside, as we moved along the isochore from low
to high temperature, we took the initial snapshot from the previous temperature point in an
attempt to converge the simulation faster by avoiding over-dissociated initial configurations.
Furthermore, the isotherms in the low pressure regime, which have a density range of 0.5 to 0.75
g/cm3, were chosen because of the unknown steepness of the IMT boundary slope at the outset of
this investigation.
Optical calculations:
The dynamic conductivity was calculated in the Kubo-Greenwood (KG) formalism [15,16] and
the dc conductivity extracted in the static field limit. All KG calculations in the main text
(BOMD and PIMD boundaries) were performed with VASP with the use of SCAN-L + rVV10
on a set of 20 evenly spaced snapshots from each MD trajectory. Note that the first 1000 steps of
each trajectory were skipped prior to taking the snapshots so as to allow for equilibration of the
simulation. The KG calculations used an automatically generated 2×2×2 Monkhorst-Pack k-
mesh with a plane-wave energy cutoff of 1000 eV and 256 bands. Results of KG convergence
tests for various thermodynamic conditions are shown in Figure S2. A potential inconsistency
may be introduced in the calculated PIMD IMT boundary as SCAN + rVV10 is used to produce
the ionic configurations and SCAN-L + rVV10 then is used to generate the Kohn-Sham orbitals
and eigenvalues for the calculated conductivity at each ionic configuration taken in the set of
snapshots. See discussion below regarding the small magnitude consequences of the
inconsistency.
The temperatures for which the average dc conductivity is directly above or below the
2000 S/cm criterion were identified. Linear interpolation between the two points then pinpointed
the transition temperature at which a dc conductivity of 2000 S/cm occurred. To estimate the
uncertainty in the calculated IMT temperature, the standard error of the dc conductivity was
added (or subtracted) from the dc conductivity for the two points in the fit and a reassessment of
-
3
the transition temperature made. This procedure yields an estimated maximum uncertainty
associated with how the IMT was calculated, of ± 30 K. However, this is not a direct measure of
uncertainty from the theoretical approximations (primarily the XC approximation).
The corresponding reflectivity was calculated from the dynamical conductivity according
to reference [17]. Figure S3 shows the dc conductivity and reflectivity at 1.96 eV (calculated
relative to vacuum) for various isochores (some of which are not shown in the main text).
SCAN vs. SCAN-L:
As noted above and in the main text, the use of SCAN in the PIMD simulations and SCAN-L in
the KG calculations introduces inconsistency in the calculated PIMD IMT boundaries. SCAN-L
has been shown to reproduce, or nearly reproduce, various structural and energetic quantities
from SCAN for solids [5] and molecules [4]. The inconsistency arises in part because of the
inequivalence of the Kohn-Sham (KS) and generalized Kohn-Sham (gKS) schemes used to
determine the ground state electronic structure [4,5]. For reasons of complexity and
computational cost, SCAN calculations use gKS. SCAN-L calculations use KS. We did several
calculations to quantify the effects of that difference.
First we calculated hydrogen dc conductivities (with QE KGEC [18]) from SCAN +
rVV10 gKS orbitals and eigenvalues on the same set of ionic snapshots from the PIMD
trajectories that we used with SCAN-L + rVV10 KS orbitals and eigenvalues in the KG
calculations with VASP. The resulting transition temperature had an average increase of 11 K.
None of the transition points had a shift greater than 20 K. This shift is smaller than the estimated
methodological uncertainty; recall discussion above. As such the difference between SCAN and
SCAN-L (and the corresponding pseudo-potentials) inputs to the KG calculations of the PIMD
boundary is inconsequential.
The second issue regarding the consistency of SCAN versus SCAN-L is the matter of the
ionic trajectories and snapshots. That is, the comparison of PIMD and BOMD IMT boundaries
includes both the direct electronic structure distinctions between SCAN and SCAN-L but also
the possible direct NQEs. For a transition boundary that is clearly dependent on the ionic
configuration set, even minor changes in its generation might have a significant impact on the
IMT temperature. We quantify that effect next.
-
4
Figure S4 shows a direct comparison of the ionic pair correlation function (PCF) from
SCAN-L + rVV10 and SCAN + rVV10 hydrogen BOMD simulations at 1.0 g/cm3 and 900 K. It
is clear that SCAN produces a system with a smaller molecular character (height of the first
peak) and molecules with a longer equilibrium bond length (position of first peak). Further
calculations of the dc conductivity along the 0.8, 0.9 and 1.0 g/cm3 isochores for BOMD
simulations are shown in Figure S5. Going from SCAN-L to SCAN to drive the BOMD
simulation shifts the IMT boundary higher in temperature by 5 to 7% for all three isochores.
Thus the shift in the IMT boundary associated with the inclusion of NQEs is underestimated by
roughly 100 K (effective cancellation of the two changes) when comparing the SCAN-L +
rVV10 BOMD results to the PIMD SCAN + rVV10 results.
PBE comparison 3000 K:
As can be seen in Figure 1 of the main text, the predicted IMT boundary from SCAN-L + rVV10
(BOMD) appears to approach that from PBE around 3000 K. To investigate this behavior
further, the dc conductivities and the height of the first peak of the PCF from PBE and SCAN-
L+rVV10 (both classical nuclei) along the 2500 K and 3000 K isotherms are compared in Figure
S6. At 3000 K, one sees clearly that the difference between PBE and SCAN-L + rVV10 results
for both properties becomes significantly smaller than at lower temperatures across the whole
density (or pressure) range of the isotherm. Furthermore, there is a striking similarity to the
results from the two functionals for the overall structure of the curves of the two properties at
3000 K that does not appear in the comparison along the 2500 K isotherms. This finding helps
support the notion that the predicted IMT boundaries of SCAN-L + rVV10 and PBE do in fact
become close above 3000 K and that the two XC functionals predict a fluid hydrogen system
with a similar level of accuracy at such thermodynamic conditions.
References:
1. G. Kresse, J. Hafner, Ab Initio molecular-dynamics simulation of the liquid-metal–
amorphous-semiconductor transition in germanium. Phys. Rev. B 49, 14251 (1994).
2. G. Kresse, J. Furthmüller, Efficiency of ab-Initio total energy calculations for metals and
semiconductors using a plane-wave basis set. Comput. Mater. Sci. 6, 1550 (1996).
3. G. Kresse, J. Furthmüller, Efficient iterative schemes for ab Initio total-energy
calculations using a plane-wave basis set. Phys. Rev. B 54, 11,16911,186 (1996).
-
5
4. D. Mejía-Rodríguez, S. B. Trickey, Deorbitalization strategies for meta-generalized-
gradient-approximation exchange-correlation functionals. Phys. Rev. A 96, 052512
(2017).
5. D. Mejía-Rodríguez, S. B. Trickey, Deorbitalized meta-GGA exchange-correlation
functionals in solids. Phys. Rev. B 98, 115161 (2018).
6. O. A. Vydrov, T. Van Voorhis, Nonlocal van der Waals density functional: The simpler
the better. J. Chem. Phys. 133, 244103 (2010).
7. R. Sabatini, T. Gorni, S. de Gironcoli, Nonlocal van der Waals density functional made
simple and efficient. Phys. Rev. B 87, 041108 (2013).
8. G. Kresse, D. Joubert, From ultrasoft pseudopotentials to the projector augmented-wave
method. Phys. Rev. B 59, 17581775 (1999).
9. M. Ceriotti, J. More, D. E. Manolopoulos, i-PI: A python interface for ab initio path
integral molecular dynamics simulations. Comput. Phys. Comm. 185, 10191026 (2014).
10. V. Kapil et al., i-PI 2.0: A universal force engine for advanced molecular simulations.
Comput. Phys. Commun. 236, 214223 (2019).
11. P. Giannozzi et al., QUANTUM ESPRESSO: A modular and open-source software
project for quantum simulations of materials. J. Phys.: Condens. Matter 21,
395502 (2009).
12. P. Giannozzi et al., Advanced capabilities for materials modelling with quantum
ESPRESSO. J. Phys.: Condens. Matter 29, 465901 (2017).
13. J. Sun, A. Ruzsinszky, J. P. Perdew, Strongly constrained and appropriately normed
semilocal density functional. Phys. Rev. Lett. 115, 036402 (2015).
14. V.-V. Karasiev, T. Sjostrom, S.-B. Trickey, Generalized-gradient approximation
noninteracting free energy functionals for orbital free density functional calculations,
Phys. Rev. B 86, 115101 (2012).
15. R. Kubo, Statistical-mechanical theory of irreversible processes. I. General theory and
simple applications to magnetic and conduction problems. J. Phys. Soc. Jpn. 12, 570586
(1957).
-
6
16. D. A. Greenwood, The Boltzmann equation in the theory of electrical conduction in
metals. Proc. Phys. Soc. Lond. 71, 585596 (1958).
17. S. X. Hu et al., Impact of first-principles properties of deuterium-tritium on inertial
confinement fusion target designs. Phys. Plasmas 22, 056304 (2015).
18. L. Calderín, V. V. Karasiev and S. B. Trickey, Kubo-Greenwood electrical conductivity
formulation and implementation for projector augmented wave datasets, Comput Phys
Commun 221, 118-142 (2017).
-
7
Fig. S1. Convergence tests performed by single point calculations for the setup of the MD
simulation parameters including the plane –wave energy cutoff, k-mesh and number of bands.
All red curves correspond to the free energy of the system (not including contribution from the
ionic entropy) and blue curves correspond to the total pressure.
-
8
Fig S2. Single snapshot calculations of the dc conductivity as a function of k points (top left and
right and bottom left) and plane-wave energy cutoff (bottom right) used to determine the
converged set of simulation parameters for the dc conductivities calculations over the set of
snapshots for each thermodynamic condition.
-
9
Fig S3. The dc conductivity (top) and reflectivity (bottom) of the PIMD isochores with the use of
SCAN-L + rVV10 in the KG calculation. Note the horizontal dotted black line in the dc
conductivity panel indicates the 2000 S/cm criterion used to determine the IMT.
-
10
Fig. S4. Pair correlation function (PCF) for 4 separate simulations of hydrogen at 1.0 g/cm3 and
900 K. Blue curve corresponds to the BOMD simulation with VASP using SCAN-L + rVV10.
Red curve is the PCF from PIMD simulation with one bead (i.e. BOMD) using SCAN + rVV10.
Green and black are PCFs from PIMD simulations with eight and four beads respectively.
-
11
Fig. S5. Comparison of dc conductivities along the 0.8 (top), 0.9 (middle) and 1.0 g/cm3
(bottom) isotherms. The dotted black line indicates a dc conductivity of 2000 S/cm. Each red
curve corresponds to the BOMD simulation with the use of SCAN-L + rVV10 and the green
curve corresponds to BOMD simulations with SCAN + rVV10. Note that the KG calculations
for both sets of data were performed with SCAN-L + rVV10.
-
12
Fig S6. Left, comparison of the dc conductivity along the 2500K and 3000K isotherms for
SCAN-L + rVV10 and PBE (BOMD). Right, first peak of the pair correlation function for the
corresponding dc conductivity curves in the left panel.
-
13
Parameter BOMD PIMD
# atoms 256 256
Plane-Wave cutoff
energy
750 eV 2700 eV
Convergence criterion
on the total electronic
free energy
1E-5 eV 2.7E-5 eV
# bands 168 160
K pt’s gamma gamma
Time step 0.1 fs 0.2 to 0.5 (adjust for
temperature and density)
# MD steps 6000 6000
Electron partial
occupancies
Fermi smearing Fermi smearing
Bare proton treatment Projector Augmented
Wavepotentials [8]
Local [14]
Ensemble NVT NVT
Thermostat Nose-Hoover (acts every 40 MD
steps)
Langevin (acts every 200
fs)
# beads N.A. 8
Table S1. Quick reference for the parameters used in the BOMD and PIMD simulations. Note,
in the case of Fermi smearing the smearing temperature is set to the ionic temperature to be
maintained by the thermostat.